Will It Run AI
CalculatorModelsHardwareCompare
Product
  • Calculator
  • Compare
  • Tier List
Browse
  • Models
  • Hardware
  • Docs
About
  • Why It Works
  • What's New
  • Legal Notice
  • Privacy Policy

All estimates are approximations based on mathematical models and public specifications. Actual performance may vary. Do not make purchasing decisions based solely on these estimates.

Data sourced from Hugging Face, Ollama, and official model documentation. Model names and logos are trademarks of their respective owners.

© 2026 Will It Run AI — Fase Consulting Ibiza, S.L. (NIF: B57969656)

Home/Hardware/GPUs/AMD Instinct MI325X 256GB
AMD

AMD

AMD Instinct MI325X 256GB

InstinctDatacenterCDNA 4OAMROCm
256GB
VRAM
6kGB/s
Bandwidth
1.3kTFLOPS
FP16 Compute
2.6kTOPS
INT8 Inference
VRAM256 GBBandwidth6k GB/sCompute1.3k TFInference2.6k TOPS
AMD Instinct MI325X 256GBCategory AvgAMD Instinct MI350X 288GB

Specifications

Compute
FP161307 TFLOPS
INT82614 TOPS
ArchitectureCDNA 4
Memory
VRAM256 GB
Bandwidth6000 GB/s
General
FamilyInstinct
SegmentDatacenter
InterconnectOAM
Compute PlatformROCM

Architecture

CDNA 4

CDNA 4 powers the next-generation Instinct MI325X and MI350X accelerators. Built on TSMC 3nm with up to 288 GB HBM3e memory and native FP4 support for maximum inference density.

AI Relevance

With up to 288 GB HBM3e and FP4 support, CDNA 4 targets the highest-density AI inference deployments. Directly competes with NVIDIA Blackwell B200 for large-scale model serving.

Process: TSMC 3nmPlatform: ROCMPrecisions: FP64, FP32, TF32, FP16, BF16, FP8, FP4, INT8

Recommendations by Workload

Agentic Coding

C

Devstral 2 123B Instruct

This model is still usable for agentic-coding, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 58.4 tok/s · 59K ctx · llama.cpp
140.0 GB / 256.0 GB VRAM

Chat

C

Mistral Small 4 119B

This model is a direct match for chat. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 175.6 tok/s · 21K ctx · llama.cpp
99.9 GB / 256.0 GB VRAM

Coding

C

Devstral 2 123B Instruct

This model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 58.4 tok/s · 34K ctx · llama.cpp
120.7 GB / 256.0 GB VRAM

RAG

C

Command A 111B

This model is a direct match for rag. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 64.7 tok/s · 64K ctx · llama.cpp
128.9 GB / 256.0 GB VRAM

Reasoning

B

Qwen 3 235B A22B

This model is a direct match for reasoning. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 81.5 tok/s · 24K ctx · llama.cpp
173.3 GB / 256.0 GB VRAM

Full Model Compatibility

AlibabaQwen 3 235B A22B
B56
235B173.3 GB82 tok/s24K ctx
moe
MistralMixtral 8x22B
C52
141B118.6 GB98 tok/s35K ctx
moe
MistralDevstral 2 123B Instruct
C51
123B120.7 GB58 tok/s34K ctx
dense
MistralMistral Small 4 119B
C51
119B100.1 GB176 tok/s41K ctx
moe
CohereCommand A 111B
C50
111B111.6 GB65 tok/s37K ctx
dense
UnslothQwen3.5 122B A10B
C50
122B105.3 GB68 tok/s39K ctx
dense
AlibabaQwen 2.5 72B
C49
72B81.7 GB100 tok/s50K ctx
dense
AlibabaQwen 2.5 VL 72B
C49
72B81.7 GB100 tok/s33K ctx
dense
MetaLlama 3.3 70B
C49
70B80.1 GB103 tok/s51K ctx
dense
AlibabaQwen3-Coder-Next
C49
80B76.1 GB272 tok/s54K ctx
moe
UnslothQwen3.5 35B A3B
C47
35B53.3 GB205 tok/s77K ctx
dense
Lmstudio-communityLQwen3.5 35B A3B
C47
35B53.3 GB205 tok/s77K ctx
dense
AlibabaQwen 2.5 Coder 32B
C46
32B51.0 GB224 tok/s80K ctx
dense
UnslothQwen3.5 27B
C46
27B47.2 GB266 tok/s87K ctx
dense
AlibabaQwen3-Coder 30B A3B Instruct
C46
30.5B45.9 GB609 tok/s89K ctx
moe
Unslothgemma 3 27b it
C46
27B47.2 GB266 tok/s87K ctx
dense
AlibabaQwen3-VL 30B A3B Instruct
C46
30B45.6 GB630 tok/s90K ctx
moe
MistralDevstral Small 2 24B Instruct
C46
24B44.9 GB299 tok/s91K ctx
dense
MistralDevstral Small 1.1
C46
24B44.9 GB299 tok/s91K ctx
dense
MistralCodestral 2 25.08
C46
22B43.4 GB326 tok/s94K ctx
dense
UnslothQwen3.5 9B
C45
9B33.4 GB798 tok/s123K ctx
dense
HauhauCSHQwen3.5 9B Uncensored HauhauCS Aggressive
C45
9B33.4 GB798 tok/s123K ctx
dense
BartowskiBMeta Llama 3.1 8B Instruct
C45
8B32.6 GB898 tok/s126K ctx
dense
XtunerXllava llama 3 8b v1 1
C45
8B32.6 GB898 tok/s126K ctx
dense
Lmstudio-communityLQwen3.5 9B
C45
9B33.4 GB798 tok/s123K ctx
dense
UnslothDeepSeek R1 0528 Qwen3 8B
C45
8B32.6 GB898 tok/s126K ctx
dense
TheBlokeTLlama 2 7B Chat
C45
7B31.9 GB1026 tok/s129K ctx
dense
MaziyarPanahiMMeta Llama 3 8B Instruct
C45
8B32.6 GB898 tok/s126K ctx
dense
TheBlokeTMistral 7B Instruct v0.2
C45
7B31.9 GB1026 tok/s129K ctx
dense
MaziyarPanahiMMistral 7B Instruct v0.3
C45
7B31.9 GB1026 tok/s129K ctx
dense
BartowskiBLlama 3.2 3B Instruct
C45
3B29.5 GB2068 tok/s139K ctx
dense
UnslothQwen3.5 4B
C45
4B29.7 GB1795 tok/s138K ctx
dense
BartowskiBgemma 2 2b it
C45
2B28.9 GB2804 tok/s142K ctx
dense
Googlegemma 2b
C45
2B28.5 GB3590 tok/s144K ctx
dense
QwenQwen2.5 3B Instruct
C45
3B29.1 GB2393 tok/s141K ctx
dense
Lmstudio-communityLgemma 3 4b it
C45
4B29.7 GB1795 tok/s138K ctx
dense
Hugging-quantsHLlama 3.2 1B Instruct Q8 0
C45
1B28.1 GB4599 tok/s146K ctx
dense
QwenQwen2.5 1.5B Instruct
C45
1.5B28.2 GB4380 tok/s145K ctx
dense
TheDrummerTGemmasutra Mini 2B v1
C44
2B28.5 GB3590 tok/s144K ctx
dense
TheBlokeTTinyLlama 1.1B Chat v1.0
C44
1.1B28.0 GB4380 tok/s146K ctx
dense
Ggml-orgGSmolVLM 500M Instruct
C44
0.5B27.7 GB4599 tok/s148K ctx
dense
Ggml-orgGembeddinggemma 300M
C44
0.3B27.5 GB4599 tok/s149K ctx
dense
MetaLlama 4 Maverick 17B 128E
C42
400B273.2 GB51 tok/s15K ctx
moe
DeepSeekDeepSeek R1 671B
F0
671B441.6 GB31 tok/s9K ctx
moe
Z.aiGLM-5
F0
744B486.6 GB28 tok/s8K ctx
moe
Moonshot AIKimi K2.5
F0
1000B641.5 GB22 tok/s6K ctx
moe
MistralMistral Large 3
F0
675B444.7 GB31 tok/s9K ctx
+1moe
AlibabaQwen3-Coder 480B A35B Instruct
F0
480B324.8 GB42 tok/s13K ctx
moe
DeepSeekDeepSeek V3 671B
F0
671B441.6 GB31 tok/s9K ctx
moe
UnslothQwen3.5 397B A17B
F0
397B330.7 GB18 tok/s12K ctx
dense

Just out of reach

Models you could run with an upgrade

High-quality models that need a bit more memory

DeepSeekDeepSeek R1 671B
671BTier 5Needs ~447.4 GB
Z.aiGLM-5
744BTier 5Needs ~492.8 GB
Moonshot AIKimi K2.5
1000BTier 5Needs ~646.5 GB
MistralMistral Large 3
675BTier 5Needs ~451.1 GB
AlibabaQwen3-Coder 480B A35B Instruct
480BTier 5Needs ~330.2 GB

Upgrade paths

Upgrade from AMD Instinct MI325X 256GB

See what you unlock with more powerful hardware

Upgrade options

Upgrade options

AMDAMD Instinct MI350X 288GBNext step up
288 GB VRAM (+32)8000 GB/s (+2000)
A
Unlocks Qwen3-Coder 480B A35B Instruct+33% faster avg

~$8,000 MSRP

Compare this GPU